Tech Lead · Platform Engineering · ML Systems
Frank Cancedda
Tech lead and platform engineer with a deep stack — Kubernetes, AI inference, and ML systems from the ground up. I own technical roadmaps, make architecture calls, and build the tooling that makes teams move faster.
What I've built
One-command dev environments
Built an orchestration platform (Tilt + k3d) that mirrors compliance-grade production. 20+ microservices — Triton, vLLM, Keycloak, Istio — with tilt up.
ML from scratch
Implemented Transformer, LLaMA, recursive TRM, and a cost-aware retrieval policy from the ground up in PyTorch. +4.9 NDCG@10 over best static hybrid baseline.
Tech lead
Owned the technical roadmap, made architecture calls, and mentored junior and senior engineers — shipping faster by making the team's work simpler and clearer.
Projects
Local k8s dev platform (Tilt + k3d) mirroring compliance-grade production. One command spins up the full AI inference stack — vLLM, Triton, Keycloak, Minio, and 15+ microservices.
Transformer architectures built from scratch: vanilla, LLaMA, recursive TRM reasoning, and a 1.6M-param cost-aware retrieval policy trained on BEIR benchmarks.
Rust game engine with WASM support and a formal API stability system. Includes program comprehension and graph-based architecture extraction tooling.
Full-stack fitness tracker with AI meal analysis, BMI tracking, and personalized goals. Live on this cluster.